| --- |
| license: agpl-3.0 |
| tags: |
| - mirid |
| - llama-cpp |
| - gguf |
| - automatic-speech-recognition |
| - text-to-speech |
| - image-text-to-text |
| - windows |
| - cuda |
| - vulkan |
| --- |
| |
| # Mirid Runtime for Windows |
|
|
| This repository hosts the frozen local-inference runtime used by the Mirid desktop app. It is a release channel, not a model. |
|
|
| On first launch, Mirid downloads the runtime archive, verifies its exact size and SHA-256 hash, then installs it in the user's application-data directory. Later Mirid releases can update this runtime independently of the desktop installer. |
|
|
| The current stable package is runtime v8. The default Windows x64 package includes: |
|
|
| - the Mirid backend and voice service, including the official Parakeet.cpp CPU transcription runtime; |
| - current `llama.cpp` runners for CPU, Vulkan, and CUDA 12; |
| - CUDA-enabled `llama-cpp-python` and `stable-diffusion-cpp-python` builds; |
| - the Python and native dependencies required by Mirid's default feature set. |
|
|
| `runtime-release.json` is the machine-readable stable-channel manifest. Mirid pins each asset by filename, byte size, and SHA-256 digest before installation. |
|
|
| ## Open models connected to Mirid |
|
|
| The linked model badges on the [Mirid Runtime Registry](https://huggingface.co/spaces/boneylizardwizard/mirid-runtime-registry) are generated from this audited set of live Hugging Face repositories. |
|
|
| - **Speech recognition:** [`openai/whisper-large-v3-turbo`](https://huggingface.co/openai/whisper-large-v3-turbo), [`UsefulSensors/moonshine-streaming-tiny`](https://huggingface.co/UsefulSensors/moonshine-streaming-tiny), [`nvidia/parakeet-tdt-0.6b-v2`](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2), [`nvidia/parakeet-tdt-0.6b-v3`](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3), [`nvidia/stt_zh_conformer_transducer_large`](https://huggingface.co/nvidia/stt_zh_conformer_transducer_large), [`nvidia/nemotron-speech-streaming-en-0.6b`](https://huggingface.co/nvidia/nemotron-speech-streaming-en-0.6b) and [`mudler/parakeet-cpp-gguf`](https://huggingface.co/mudler/parakeet-cpp-gguf). |
| - **Text to speech and voice conversion:** [`ResembleAI/chatterbox`](https://huggingface.co/ResembleAI/chatterbox), [`ResembleAI/chatterbox-turbo`](https://huggingface.co/ResembleAI/chatterbox-turbo), [`ResembleAI/chatterbox-nano`](https://huggingface.co/ResembleAI/chatterbox-nano), [`openbmb/VoxCPM2`](https://huggingface.co/openbmb/VoxCPM2) and [`hexgrad/Kokoro-82M`](https://huggingface.co/hexgrad/Kokoro-82M). |
| - **Vision and structured image reading:** [`LiquidAI/LFM2.5-VL-450M`](https://huggingface.co/LiquidAI/LFM2.5-VL-450M), [`LiquidAI/LFM2.5-VL-450M-Extract`](https://huggingface.co/LiquidAI/LFM2.5-VL-450M-Extract), [`LiquidAI/LFM2.5-VL-1.6B`](https://huggingface.co/LiquidAI/LFM2.5-VL-1.6B), [`LiquidAI/LFM2.5-VL-1.6B-Extract`](https://huggingface.co/LiquidAI/LFM2.5-VL-1.6B-Extract) and [`google/gemma-3-4b-it`](https://huggingface.co/google/gemma-3-4b-it). |
|
|
| ## Hugging Face plumbing |
|
|
| Behind those model integrations, Mirid uses the Hub API for repository metadata and verified downloads. These are implementation details, not part of the linked model catalogue. |
|
|
| Model availability, licences and access conditions are governed by each model repository. A repository named here is an integration point, not a claim that its weights are included in every runtime download. |
|
|
| Source code is available in the [Mirid source repository](https://github.com/boneylizard/Eloquent) under the GNU Affero General Public License v3. The repository URL still carries the project's former name. Bundled third-party components remain subject to their respective licences. |
|
|
